A space of your own.
Private by design. Your conversations, files and memory in an environment you control, with a clear view of what stays there and what leaves.
Privacy begins with ownership.Early development · Private web pilot
A home for
your assistant.
On your infrastructure.
I'm building Radhouse around one private web conversation with your assistant. Close the browser, come back later and pick up where you left off.
One conversation, over timeTry it below ↙
A conversation you can return to.
A notebook holds your question about a small garden. The day begins, and your private conversation has a place to stay.
A simple place to begin.
Open the private web app and ask your assistant a question. One conversation, with room to think.
The reply joins the conversation.
Your message and the assistant's reply stay together, ready for you to pick up again.
You can step away.
Close the browser. Your earlier messages and the assistant's context are kept.
Pick up the same conversation.
Sign in again when needed and continue where you left off, with the same assistant and earlier context.
An interactive illustration. No message is sent.
What I'm building toward
Always there.
Still yours.
A private, always-on agent that gets to know you, carries useful work forward and lives on infrastructure you control.
Personal agents like Grok Bot, Meta's Muse and OpenAI's Dots point toward this kind of relationship with AI. Radhouse is my open-source take on that idea, built around security, privacy and explicit boundaries.
Power, with clear limits.
A separate space for the agent, with limits on files, tools and network access enforced outside its prompts. Clear approval for consequential actions. Permissions you can inspect and revoke.
Security is part of the design.Open to make your own.
Open-source code to inspect, run and adapt. Room to choose models and tools, understand how they connect and shape an agent around your life.
Built in the openThe ambition: useful work can continue while you step away, within the access you've chosen. These are the principles guiding the build.
A smaller beginning
One good conversation.
I'm starting with one assistant and one private web conversation you can return to. Everyday use will help decide what deserves to grow from there.
The build, as it stands
A private pilot.
A conversation that stays with you, files you can return to and a browser you can use together. Built around one Hermes assistant in your own space.
Pilot notes & what's next
Ask about an attached PDF or document, then follow up. Saved originals remain available for the assistant to read the parts it needs, with source references, instead of putting the whole file into every prompt.
Library keeps your shared files close. Browser opens on demand: watch the assistant, take the controls, then return them when its reply has finished. Saved logins stay in the private browser's credential vault, with explicit save and remove controls.
This is a private pilot, with each new capability checked before activation. Scanned-document OCR, audio transcription, streamed reply text and a reliable Stop action still need qualification. Hosting Radhouse and choosing a model are separate decisions: data sent to a connected model follows that provider's handling.
Read the pilot build Follow the latest workNow in the private pilot
Browse together.
Take the controls.
Open a browser beside your conversation, even before your first message. It's the same session your assistant uses, so you can see the page and pick up the controls when you need to.
Navigate yourself, fill a form or sign in. Return control when the assistant has finished its reply. Share a page into the conversation when you want to talk about it.
Open when you need it. Close when you are done. Hiding the view keeps the browser open; Close ends the session.
Browser previewHide view Show view
google.comRunning Digitally
- 01 Open browser
- 02 Take control
- 03 Return to agent
Google search for Running Digitally. Hide/Show controls this preview.
Make it yours
Your context.
Your choices.
Know what your assistant remembers. Choose how you work. Explore a character that makes the space feel like yours.
In the private pilot
01 / Memory
About You.
Read the notes your assistant has saved about you and its ongoing memory. See the saved context in one place; editing those notes comes later.
02 / Your space
A terminal of your own.
Open a shell when you need it. Choose whether to include a short excerpt in your next message. Terminal context starts off; you decide when to share it.
03 / The next reply
Choose the model.
Pick from the models offered by your connected engine. Thinking choices appear only when supported. Each choice belongs to that message.
Available in the private pilot since 8 October. Choices depend on your connected model engine. Follow the release
Visuals · Design preview
A familiar face.
Three worlds to explore. One assistant to make your own.
Hearthside — meet all 4 characters
Kiln Club — meet all 3 characters
Signal Station — meet all 3 characters
Built in the open
Help shape
the next version.
What would you want an agent to remember, take care of or ask you before doing? Those are the questions I want to build around.
If you're exploring the same questions, follow the build or bring an idea to the repository.
Explore the current build Share an idea